Machine-learning - the way of the future?
Updated: May 28, 2020
We all understand how amazing technology is and how rapidly it is evolving and morphing into so many useful forms. From a device that can vacuum the house (with the cat sitting on it, and zooming around) with just the click of a button, to iris and/or face real-time recognition, like it or not, technology and its advances are part of our everyday lives. It doesn’t stop there though, this growth (malignant or not) has rapidly metastasized to involve other spheres of our lives, and probably most controversially (and thus most relevant to our blog), human and more recently, veterinary medicine.
Plus, what would this website and blog be without a post on something virtual and/or technology-based.
Firstly, what is it? Recognized as a sub-field of artificial intelligence, machine-learning is aimed at utilizing, recognizing, and learning from past behaviors and/or rules of old data, so as to forecast future behaviors, outcomes, etc. In other words, its the equivalent of a well-thought through and (actually) carried out New Years resolution, something that many of us strive to achieve each year, yet somehow always seem to fall short. Machine-learning has been applied in many fields, from banking, to education, through to medicine, security, telecommunication, and yes, even veterinary science.
Secondly, can it help us, and if so, how? Well, if you weren’t convinced by the mutterings above, let‘s see if we can change that. Probably one of the easiest ways to explain its use is with the recent paper by Spiteri et al. (2019). This group of authors looked at the use of machine learning to understand neuromorphological change and image-based biomarker identification in Cavalier King Charles Spaniels with Chiari-like malformation-associated pain and syringomyelia. Ok, yes, a big mouthful to ruminate on, but given how complex and to some extent, poorly understood this disease process and its clinical manifestations are, it seems fitting to describe it with such terms.
A particular point raised in this paper, that I thought was worth drawing your attention to, is the following:
"Traditional morphometric studies also have the potential disadvantage that they are hypothesis-driven (ie, the researcher proposes the measurement to be tested based on prior knowledge or intuition)."
This statement is contrasting a hypothesis-driven approach (more traditional means of formulating a study), with that of machine-learning, which is considered to be a purely data-driven approach. It serves to highlight the inherent (and unavoidable) bias, that is present with many studies today, formulated on the basis of an individual's knowledge, experience, and/or intuition, which very well may be the exception, rather than the norm. I won't poke the bear any more with this line of thought though.
Another statement from the article, because the authors summed it up so nicely...
"In other words, we propose to let the data “speak” in an unbiased way."
Granted, although machine-learning may not be practical and/or even justified for several scopes of veterinary science (even particular disease processes), the argument for its use with previously poorly-understood diseases, does seem valid. The reason being, as eluded to above, is that machine-learning can not only process a considerable volume of data, but more importantly, with the application of certain algorithms, process such data in an unbiased way. This facilitates recognition of key factors, that could have been missed if a more conventional hypothesis-driven approach was applied. That being said, (insert limitations section...) further studies are required to truly contrast one method versus the other, in terms of both practicality and clinical usefulness.
Thirdly, where has it been used so far, and has it been successful? In human medicine, machine-learning has been trialed and successfully made use of in the following cases/fields/sub-specialties, etc. (to mention only a few):
Automated blood smear analysis
Breast cancer diagnosis, prognosis, and predicting recurrence
Detection of subjects and brain regions related to Alzheimer’s disease
Predicting multiple sclerosis disease course
In veterinary medicine, although lagging behind somewhat, several studies using machine-learning can be found in the veterinary literature. Those directly pertaining to small animals, include the following:
Canine pelvic radiographs
Chiari-like malformation-associated pain and syringomyelia in Cavalier King Charles Spaniel
Classification of radiographic lung patterns
Distinguishing between inflammatory bowel disease and alimentary lymphoma in cats
Predicting early risk of chronic kidney disease in cats
Other fields of veterinary science have also cottoned on to the use of this modern modality, with some examples worth mentioning including, but not limited to the following:
Detection of lameness in sheep
Lameness scoring system for dairy cows using force plates and artificial intelligence
Porcine breeding herds
Precision animal agriculture
Prediction of conception success to a given insemination in lactating dairy cows
Future directions and where machine-learning might spread to next - well, where do we start:
'Machine Learning Algorithm as a Diagnostic Tool for Hypoadrenocorticism in Dogs' was presented as a research abstract at the ACVIM Forum 2019, by Kryste Reagan, which showed promising results...see below:
"The MLA had a sensitivity of 96.3% (95% CI, 81.7-99.8%), specificity of 97.2% (95% CI, 93.7-98.8%) and overall accuracy of 99.4%, greater than that of Na/K ratio, total lymphocyte count and a logistic regression combining the Na/K and lymphocyte count."
Other diseases that we suspect may soon be targeted with machine-learning include meningoencephalomyelitis of unknown etiology (MUE), other immune-mediated disease processes, and potentially even diabetes mellitus management and/or risk for diabetic ketoacidosis - watch this space!
Some useful links and light-reads:
Awaysheh A, Wilcke J, Elvinger F, et al. (2019), Review of Medical Decision Support and Machine-Learning Methods. Vet Pathol, 56(4):512-525.
Bradley R, Tagkopoulos I, Kim M, et al. (2019), Predicting early risk of chronic kidney disease in cats using routine clinical laboratory tests and machine learning. J Vet Intern Med, 33:2644–2656. Available online: https://onlinelibrary.wiley.com/doi/full/10.1111/jvim.15623
Nie A, Zehnder A, Page RL, et al. (2018), DeepTag: inferring diagnoses from veterinary clinical notes. NPJ Digit Med, 1:60. Available online: https://www.nature.com/articles/s41746-018-0067-8
Spiteri M, Knowler SP, Rusbridge C, Wells K. (2019), Using machine learning to understand neuromorphological change and image-based biomarker identification in Cavalier King Charles Spaniels with Chiari-like malformation-associated pain and syringomyelia. J Vet Intern Med, 33:2665–2674. Available online: https://onlinelibrary.wiley.com/doi/full/10.1111/jvim.15621